• Skand Pty. Ltd.

    Building Envelope Inspection Powered by Machine Learning and Reality Modeling for RMIT University Brunswick Campus

    Victoria, Australia

Project Summary

Project

Skand Pty. Ltd., a Melbourne-based start-up, was tasked with conducting an urban capture of the Royal Melbourne Institute of Technology’s (RMIT) Brunswick campus. The Australian public research university wanted to integrate drone data capturing and analysis into its award-winning 40-year asset lifecycle program, which is ISO certified. To better manage the assets, RMIT wanted to create a digital version of its Brunswick campus, updating the previous 2D orthophoto representation. The AUD 150,000 project area covers 6.5 hectares and six buildings of various sizes and building material types, which would all be incorporated into the RMIT asset lifecycle program. Skand needed to conduct an unbiased analysis of the site that included layers of categorization and integrated with an ISO-compliant asset management system.

Solution

Skand used a Phantom 4 Pro drone to capture imagery of the project site, conducting three passes with the drone. Each pass would be closer to the asset until it was as low as legally possible. Skand then used its web application, supported by ContextCapture, to integrate the drone imagery and analysis into the ISO-certified asset lifecycle program. Skand was able to take in aerial- and land-based 2D images and turn them into meaningful data sets mapped in a 3D reality mesh. This mesh was then published online, detailing condition and priority for any required action or intervention. Using machine learning, the project team ran analyses over the drone-captured images to determine whether the image set was of sufficient quality to be processed. Once it was deemed a good quality, the team processed the images to stitch the 2D images together into a 3D surface.

Outcome

By using drones and ContextCapture, Skand realized significant time and cost savings. Spreading over 6.5 hectares, the Brunswick campus’ building envelope inspection would normally have taken at least a month. However, this project’s drone imagery capture took three and a half days, and the processing time for the model only took two days, allowing the project to occur nearly four times quicker than with the photogrammetry engine previously used on similar projects. Skand delivered an inspection report with a 3D campus model into the web platform within seven days at a cost of AUD 80,000, which was AUD 70,000 less than the estimated budget. Skand expects that the processing time will be reduced to just 72 hours within the next six months by recycling images that failed initial processing with another round of supervised machine learning and testing the solution algorithms. This solution also improves safety, as roof and façade inspections no longer require colleagues to leave the ground. It is faster because works can be automatically carried out on computers, and it is 60% cheaper than traditional inspection methods. The analysis is consistent across regions, making it more accurate.

Software

ContextCapture’s quality of outputs and accuracy of mapping defects was far greater when compared to previous outputs modeled in different photogrammetry engines. ContextCapture was used to support Skand’s web application for this project. ContextCapture also allowed Skand to create the 3D reality mesh based on 2D images. The application let Skand use thermal imagery to detect not only thermal signatures of mechanical-electrical assets, but also water leaks into roof insulation methods. Using artificial intelligence (AI) within ContextCapture, Skand created 16 categories of defects on both the roofs and facades, including cracks, moss, algae, rush, and other forms of corrosion and building material degradation. Skand’s web application displays the ContextCapture 3D reality mesh with the detected defects mapped out, creating an inspection and condition report powered by Power BI.

Project Playbook: ContextCapture

Outcome/Facts
  • Skand was tasked with conducting an urban capture of RMIT’s Brunswick campus, supported by ContextCapture, to integrate drone imagery and analysis into an ISO-certified asset lifecycle program.
  • ContextCapture allowed building envelope inspection to occur nearly four times faster than with previously used photogrammetry engines.
  • Skand delivered an inspection report and 3D campus model within 7 days at AUD 80,000, significantly less than the AUD 150,000 budget.